Classification of contagious lung infections is complicated due to similar clinical manifestations and overlapping computed tomography imaging features. Federated deep learning – based disease detection models are being increasingly utilized to secure patient data sourced from multiple medical centers. The key aspiration of this study is to investigate the potency of the federated deep neural model optimized by the Canis-collie algorithm for classification of tuberculosis, pneumonia, and COVID-19 infections. A significant highlight of this research is the chest computed tomography dataset based on prominent imaging patterns associated with the afore mentioned lung infections. The dataset is compiled by experienced pulmonologist and radiologists of Symbiosis University Hospital and Research Center, Pune, India. When tested on the curated dataset, the federated deep model achieved maximum accuracy, recall, specificity, and precision of 93.02%, 92.98%, 93.12%, and, 91.77%, respectively. The results obtained prove the efficacy of federated learning and bio-optimizers in a complex classification scenario.
